Corticosteroids compared with no corticosteroids for treatment of acute respiratory distress syndrome within 14 days of diagnosis

ARDS: acute respiratory distress syndrome; GRADE: Grading of Recommendations Assessment, Development and Evaluation; CI: confidence interval; RCT: randomized controlled trial; RR: risk ratio; MD: mean difference; ICU: intensive care unit; M-H: Mantel-Haenszel; df: degrees of freedom; SD: standard deviation; IV: inverse variance; CAP: community-acquired pneumonia.
* The effect estimates for this outcome are based on the pooled effect for patients who received steroids before day 14 in trials that were not stopped early for benefit (panel D). We detected a subgroup effect for patients randomized before day 14 versus on or after day 14 (panel C), so we restricted our analysis to patients randomized within 14 days. We did not detect a significant subgroup effect on mortality in trials stopped early for benefit versus not stopped early for benefit (panel D); however, we did detect a significant subgroup effect for this when assessing ventilator-free days (panel E). We decided to trust the more conservative estimate based on the 7 trials that were not stopped early for benefit.
¶ We rated down for the following concerns:

1. Indirectness in the outcome: The duration of follow-up in 2 trials was relatively short (<15 days)^[6,7].
2. Indirectness in the patient population: We are interested in the effect of steroids in patients with ARDS managed with low tidal volume ventilation. Only 3 trials specifically described use of low tidal volume ventilation^[8-10], 1 trial described using larger tidal volumes^[1], and 3 trials did not provide details regarding tidal volumes. In addition, 2 trials enrolled only patients with severe sepsis^[1,9] and 1 trial included only patients with CAP^[27]; it's possible that the effect of steroids in these trials is due to effects on CAP or septic shock rather than directly on ARDS.
3. Risk of bias: 2 trials provided only limited details regarding randomization and allocation concealment^[6,7], and we had limited information on the methods of a third trial because it was not published in English^[3]. A previous meta-analysis^[11] did not detect a subgroup effect based on risk of bias but did report a statistically significant subgroup effect when comparing smaller trials (<60 patients) versus larger trials (≥60 patients), with a larger effect size in smaller trials.

While each concern alone was not serious enough to warrant rating down, taken together, they may limit the certainty of our findings. We therefore rated the certainty of the evidence as moderate to high for this outcome.
Δ We decided to trust the more conservative estimate based on the 6 trials that were not stopped early for benefit (panel E).
◊ The 95% CI for the absolute effect includes possible harm with steroids.
§ We decided to trust the more conservative estimate based on the 2 trials that were not stopped early for benefit (panel F).
¥ The 95% CI for the absolute effect approaches no effect.
‡ The 95% CI for the absolute effect varies broadly from a considerable reduction to a considerable increase in neuromuscular weakness with steroids.
† Defined as blood glucose >150 mg/dL or requiring insulin therapy.